The repo contains volume element potential model, example Python code for our manuscript (under review).
Parameter-Free and Electron Counting Satisfied Representation for Machine Learning Potential Energy and Force Fields Bin Xi, Man Kit Chan, Kejie Bao, Wenjing Zhao, Ho Ming Chan, Hang Chen, and Junyi Zhu
A new material representation (called volume element) is proposed to describe materials (hexagonal-BN and Si systems in this paper) for a training of neural network potentials and force feilds. Such represention uses geometric information of the target material system and is parameter efficient, as well as satisfiying electron counting model. In addition, an efficient reference data construction scheme is proposed to generate dataset with countinous and relative even populations by appling perturbations on ground-state configuration. Such perturbation-based scheme contains stable, meta-stable, and unstable configurations without bias.
TBC...
Our paper is now under review, and the source code will be uploaded once is paper is published.